Remote Sensing Image Detection Based on Improved RT-DETR

Detecting accurately targets in remote sensing imagery is increasingly important for consumer electronics applications, such as drone-based surveying, autonomous vehicle navigation, and smart agriculture monitoring. However, real-time detection on embedded CE platforms is challenged by small target sizes, cluttered backgrounds, and limited computational resources. To tackle these challenges, we present a lightweight yet accurate detection model based on the RT-DETR framework, designed for efficient deployment on CE devices. Our improvements include a more efficient backbone network that preserves small-target semantics, a multi-path attention aggregation module for robust feature fusion, and an auxiliary feature enhancement module. Evaluated on the VisDrone2019 dataset, our model achieves 51.5% mAP@0.5, outperforming the original RT-DETR by 3.5% while reducing parameters by 16.9%, demonstrating its suitability for real-time CE applications.